System and method for analyzing cortex thickness map of hypoxic ischemic brain injury

Abstract

The invention relates to the technical field of medical image processing, in particular to a cortical thickness atlas analysis system and method for hypoxic ischemic brain injury, and the system comprises a cortical thickness data acquisition unit which generates a cortical thickness thermodynamic diagram with the precision of 0.1 mm through a Laplacian manifold reconstruction algorithm; the fNIRS blood flow data acquisition unit is used for acquiring blood flow signals of 1024 channels and generating a blood flow signal thermodynamic diagram; the multi-center data normalization unit is used for constructing and normalizing multi-center data, and performing Z-score; the part 3 is judged to be abnormal; the sensitive value selection unit is used for superposing the skin layer thinning area and the blood oxygen signal abnormal area on the basis of a topological space intersection mapping technology, and determining a sensitive value 0.9 and an injury responsibility area; and the responsibility area generation unit is used for accurately positioning a responsibility area boundary, constructing a multi-dimensional feature space and generating a responsibility area map, so that high-precision fusion of structural and functional data is realized.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a system and method for analyzing cortical thickness atlases in hypoxic-ischemic brain injury, used to accurately locate and assess the area of ​​responsibility for cortical injury in children with hypoxic-ischemic brain injury. Background Technology

[0002] Hypoxic-ischemic brain injury (HIBE) is a common type of brain injury in newborns and infants, leading to severe neurodevelopmental disorders. Accurate localization and assessment of the area of ​​responsibility for brain injury are crucial for early intervention and prognostic evaluation. Currently, clinically used assessment methods primarily rely on single-modality imaging examinations, such as magnetic resonance imaging (MRI) or functional near-infrared spectroscopy (fNIRS). However, single-modality imaging cannot simultaneously reflect structural changes and functional abnormalities in brain tissue, resulting in inaccurate localization of the area of ​​responsibility and impacting clinical decision-making.

[0003] Traditional MRI analysis methods can detect changes in cortical thickness, but they lack sensitivity to minute sub-millimeter changes and cannot assess functional status. While fNIRS can provide information on blood oxygenation-dependent levels, its spatial resolution is limited, making it difficult to accurately map to anatomical structures. Furthermore, the current lack of effective multimodal data fusion methods hinders the collaborative analysis of structural and functional information. Additionally, the lack of standardized evaluation systems makes it difficult to compare results from different centers, limiting the integrated use of multi-center data.

[0004] Therefore, there is an urgent need to develop an analysis system that can integrate structural and functional information to accurately locate the area of ​​responsibility for brain injury, providing more precise guidance for clinical diagnosis and treatment. Summary of the Invention

[0005] The purpose of this invention is to provide a system and method for analyzing cortical thickness atlases in hypoxic-ischemic brain injury, which achieves precise localization and assessment of the brain injury responsibility area by fusing high-precision structural imaging (3D-MRI) and functional imaging (fNIRS) data.

[0006] This invention proposes a cortical thickness atlas analysis system for hypoxic-ischemic brain injury, comprising:

[0007] The cortical thickness data acquisition unit is used to acquire 3D-MRI and perform cortical reconstruction. It automatically measures the cortical thickness using the Laplacian manifold reconstruction algorithm and generates a cortical thickness heatmap with an accuracy of 0.1 mm.

[0008] The fNIRS blood flow data acquisition unit is used to acquire fNIRS blood flow signals from the cerebral cortex, preprocess them based on blood oxygen dependence levels, and generate a blood flow signal heatmap.

[0009] A multicenter data normalization unit is connected to the cortical thickness data acquisition unit and the fNIRS blood flow data acquisition unit. It is used to construct multicenter data using cortical thickness data and analyze the multicenter data. It supports the normalization of multicenter data and judges the part with Z-score > 3 as abnormal.

[0010] The sensitivity value selection unit, connected to the multi-center data normalization unit, is used to superimpose the cortical thinning region and the abnormal blood oxygen signal region based on the topological space intersection mapping technology, select the intersection region of the two as the damage responsibility area, determine the sensitivity value of 0.9 through sensitivity value analysis, and finally determine the damage responsibility area.

[0011] The responsibility area generation unit, connected to the sensitivity value selection unit, is used to generate a responsibility area map and responsibility area sensitivity values ​​based on the abnormal conditions of the cortex.

[0012] Preferably, the sensitivity value selection unit includes:

[0013] The topology space construction module is used to convert cortical thickness data into structural topology space, blood oxygenation level data into functional topology space, and establish a homeomorphic mapping relationship between the two spaces.

[0014] The spatial registration module, connected to the topological space construction module, is used to establish a unified brain standard spatial coordinate system, apply an elastic registration method to align two topological spaces, construct a point-to-point mapping relationship, and control the accuracy within 0.5mm.

[0015] The mapping quality assessment module, connected to the spatial registration module, is used to design a mapping quality assessment index system to evaluate point correspondence accuracy, regional preservation, and topological consistency, and to iteratively optimize mapping parameters.

[0016] Preferably, the sensitivity value selection unit further includes:

[0017] The intersection region extraction module is used to calculate the overlap between the cortical thinning region and the blood oxygenation abnormality region in a unified coordinate space, apply spatial correlation analysis to identify significantly related regions, and construct a preliminary intersection region representation;

[0018] The distance measurement module, connected to the intersection region extraction module, is used to calculate the Hausdorff distance between regions, quantify region similarity, and design a partitioning calculation strategy to handle irregularly shaped regions.

[0019] The intersection region optimization module, connected to the distance metric module, is used to apply a region growing algorithm to optimize the intersection boundary, introduce morphological processing to eliminate noise and small holes, and construct a hierarchical representation of the intersection region.

[0020] Preferably, the sensitivity value selection unit further includes:

[0021] The sensitive value candidate set generation module is used to determine the initial range of sensitive values ​​[0.7, 0.95] based on historical data analysis, and to design a uniform sampling strategy to generate a sensitive value candidate set;

[0022] The prior knowledge modeling module, connected to the sensitivity value candidate set generation module, is used to integrate clinical expert knowledge to construct a prior distribution and adjust the prior considering factors such as patient age and injury type.

[0023] The posterior probability calculation module, connected to the prior knowledge modeling module, is used to update the probability distribution of the sensitivity value based on the observed data, apply an iterative optimization algorithm to maximize the posterior probability, verify the consistency of the sensitivity value on different subsets, and finally determine the sensitivity value of 0.9 as the optimal threshold.

[0024] Preferably, the responsibility area generation unit includes:

[0025] The boundary precision positioning module is used to enhance the boundary of the responsibility area by applying edge detection algorithms, introduce multi-directional gradient analysis to improve boundary continuity, apply sub-pixel boundary positioning technology to improve accuracy, and generate a high-precision boundary representation with an accuracy of 0.2mm.

[0026] A hierarchical boundary representation module, connected to the boundary precise positioning module, is used to construct a multi-level boundary representation of the responsibility area, define the core area, transition area and edge area, and assign reliability weights to different level areas.

[0027] The boundary characteristic analysis module, connected to the layered boundary representation module, is used to analyze the geometric characteristics and curvature distribution of the boundary, identify the feature points and key segments of the boundary, and evaluate the relationship between the boundary and the anatomical structure.

[0028] Preferably, the cortical thickness data acquisition unit includes:

[0029] The 3D-MRI acquisition module is used to acquire three-dimensional structural images of the skull;

[0030] The preprocessing module, connected to the 3D-MRI acquisition module, is used to remove the effects of skull, dura mater, etc. from the acquired 3D-MRI.

[0031] A cortical reconstruction module, connected to the preprocessing module, is used to automatically measure cortical thickness using the Laplacian manifold reconstruction algorithm;

[0032] A heat map generation module, connected to the cortical reconstruction module, is used to generate a cortical thickness heat map with an accuracy of 0.1 mm.

[0033] Preferably, the fNIRS blood flow data acquisition unit includes:

[0034] The signal acquisition module is used to measure 1024 channels of skin blood flow signals in the brain of the child through an fNIRS helmet when the child is undergoing resting-state fNIRS blood flow signal acquisition.

[0035] The coordinate resampling module, connected to the signal acquisition module, is used to resample the signal in the brain structure coordinate system and use the cortical reconstruction data to perform spatial matching of blood flow data and calculate the blood flow signal on the cortex.

[0036] The signal preprocessing module, connected to the coordinate resampling module, is used to preprocess the blood flow signal, apply a sliding window filtering algorithm to smooth periodic noise, calculate the blood oxygen dependence level, and mark the area with a blood oxygen dependence change rate of less than -10% as the damage area.

[0037] Preferably, the multi-center data normalization unit includes:

[0038] The data acquisition module is used to acquire data on cortical thinning and decreased blood oxygenation.

[0039] A region segmentation module, connected to the data acquisition module, is used to segment the thinned cortex according to the cortex thickness map;

[0040] An anomaly detection module, connected to the region segmentation module, is used to divide the range of brain structural abnormalities according to the standard threshold Z-score > 3 and mark these points;

[0041] The consistency analysis module, connected to the anomaly detection module, is used to analyze all data, achieve consistency of multi-center data, and establish a unified cortical thickness map analysis system.

[0042] The damage area generation module, connected to the consistency analysis module, is used to generate the cortical damage area.

[0043] Preferably, the responsibility area generation unit further includes:

[0044] The dynamic characteristic analysis module is used to analyze the temporal dynamic characteristics of blood oxygenation signals within the area of ​​responsibility, extract the frequency, amplitude, and phase features of the signals, and identify abnormal dynamic patterns.

[0045] A multi-dimensional feature space construction module, connected to the dynamic characteristic analysis module, is used to integrate structure, function and dynamic characteristics, construct a multi-dimensional feature space, design a feature normalization strategy, and construct a responsibility region feature fingerprint.

[0046] The responsibility zone classification module is connected to the multi-dimensional feature space construction module. It is used to design a responsibility zone severity assessment system, classify responsibilities based on feature fingerprints, and build a graded standard library.

[0047] The responsibility area visualization module, connected to the responsibility area hierarchical classification module, is used to design a multimodal fusion visualization scheme, using color coding to express different characteristics and constructing an interactive visualization interface.

[0048] A method for analyzing cortical thickness maps in hypoxic-ischemic brain injury includes the following steps:

[0049] 3D-MRI was acquired and cortical reconstruction was performed. Cortical thickness was automatically measured using the Laplacian manifold reconstruction algorithm, generating a cortical thickness heatmap with an accuracy of 0.1 mm.

[0050] fNIRS blood flow signals of the cerebral cortex were acquired. Skin blood flow signals of 1024 channels were measured in the brain of the child using an fNIRS helmet. The signals were resampled in the brain structural coordinate system, and spatial matching of blood flow data was performed using cortical reconstruction data. Blood flow signals on the cortex were calculated. Blood flow signals were preprocessed, and a sliding window filtering algorithm was applied to smooth periodic noise. Blood oxygen dependence level was calculated, and areas with blood oxygen dependence change rate less than -10% were marked as lesion areas.

[0051] Using cortical thickness data, a multicenter dataset is constructed and analyzed. The normalization of the multicenter dataset is supported, and parts with Z-score > 3 are judged as anomalies.

[0052] Based on the topological space intersection mapping technique, the thinning cortical region and the abnormal blood oxygenation signal region are superimposed. The cortical thickness data is converted into structural topological space, and the blood oxygenation level data is converted into functional topological space. A homeomorphic mapping relationship is established between the two spaces. The overlap between the thinning cortical region and the abnormal blood oxygenation region is calculated in a unified coordinate space. Spatial correlation analysis is applied to identify significantly related regions. The Hausdorff distance between regions is calculated to quantify the similarity between regions. The intersection region between the two is selected as the damage responsibility area. The sensitivity value is determined to be 0.9 through sensitivity value analysis, and the damage responsibility area is finally determined.

[0053] Based on the abnormalities in the cortex, edge detection algorithms are applied to enhance the boundaries of the responsibility area, construct a multi-level boundary representation of the responsibility area, analyze the geometric characteristics and curvature distribution of the boundary, analyze the temporal dynamic characteristics of blood oxygenation signals within the responsibility area, integrate structural, functional and dynamic characteristics, construct a multi-dimensional feature space, and generate a responsibility area map and responsibility area sensitivity values.

[0054] The beneficial effects of this invention include:

[0055] 1. Achieved high-precision fusion of structural and functional data: Through topological space intersection mapping technology, the cortical thickness heatmap with a precision of 0.1mm was accurately registered and superimposed with the blood flow signal heatmap, solving the problem of inaccurate structure-function correspondence in traditional methods.

[0056] 2. Improved accuracy of responsibility area positioning: By mapping the intersection of abnormal regions in multidimensional topological space and combining a sensitivity value determination mechanism based on probability statistics, the accuracy of responsibility area positioning is increased from 72% in traditional methods to 91%, and the boundary positioning accuracy reaches 0.2mm.

[0057] 3. Enhanced objectivity and consistency of evaluation: By normalizing multi-center data and setting a standardized threshold (Z-score>3), consistency analysis of data from different centers was achieved, supporting multi-center collaborative research and result comparison.

[0058] 4. Improved clinical decision support: Through precise characterization of the responsibility area and dynamic characteristic analysis, more accurate assessment indicators are provided for clinicians, improving the targeting of treatment plans by 62% and the accuracy of prognosis prediction by 48%.

[0059] 5. Improved processing efficiency: The system processing time has been reduced from 45 minutes in the traditional method to 8 minutes, and the computing resource requirements have been reduced by 35%, supporting efficient clinical applications. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the overall architecture of the hypoxic-ischemic brain injury cortical thickness mapping analysis system of the present invention;

[0061] Figure 2 This is a schematic diagram of the structure of the cortical thickness data acquisition unit of the present invention;

[0062] Figure 3 This is a schematic diagram of the fNIRS blood flow data acquisition unit of the present invention;

[0063] Figure 4 This is a schematic diagram of the structure of the multi-center data normalization unit of the present invention;

[0064] Figure 5 This is a schematic diagram of the sensitive value selection unit of the present invention;

[0065] Figure 6 This is a schematic diagram of the structure of the responsibility area generation unit of the present invention;

[0066] Figure 7 This is a flowchart illustrating the topological space intersection mapping technique of the present invention;

[0067] Figure 8 This is a schematic diagram of the sensitivity value determination process of the present invention;

[0068] Figure 9 This is a schematic diagram illustrating the precise positioning of the boundary of the area of ​​responsibility of this invention;

[0069] Figure 10 This is a flowchart of the method for analyzing cortical thickness atlases in hypoxic-ischemic brain injury according to the present invention. Detailed Implementation

[0070] Please refer to Figures 1-10 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0071] like Figure 1 As shown, the hypoxic-ischemic brain injury cortical thickness atlas analysis system of the present invention includes: cortical thickness data acquisition unit 1, fNIRS blood flow data acquisition unit 2, multicenter data normalization unit 3, sensitivity value selection unit 4, and responsibility area generation unit 5.

[0072] The cortical thickness data acquisition unit 1 acquires 3D-MRI data and performs cortical reconstruction. It automatically measures cortical thickness using the Laplacian manifold reconstruction algorithm, generating a cortical thickness heatmap with an accuracy of 0.1 mm. The fNIRS blood flow data acquisition unit 2 acquires fNIRS blood flow signals from the cerebral cortex and preprocesses them based on blood oxygen dependence levels to generate a blood flow signal heatmap. The multicenter data normalization unit 3 is connected to the cortical thickness data acquisition unit 1 and the fNIRS blood flow data acquisition unit 2. It uses cortical thickness data to construct multicenter data, analyzes the multicenter data, supports the normalization of multicenter data, and identifies regions with a Z-score > 3 as abnormal. The sensitivity value selection unit 4 is connected to the multicenter data normalization unit 3. Based on topological spatial intersection mapping technology, it overlays thinned cortical regions with abnormal blood oxygenation signal regions, selects the intersection region as the injury responsibility area, and determines a sensitivity value of 0.9 through sensitivity value analysis, ultimately identifying the injury responsibility area. The responsibility area generation unit 5 is connected to the sensitivity value selection unit 4 and is used to generate a responsibility area map and responsibility area sensitivity values ​​based on the abnormal conditions of the cortex.

[0073] like Figure 2 As shown, the cortical thickness data acquisition unit 1 includes: a 3D-MRI acquisition module 11, a preprocessing module 12, a cortical reconstruction module 13, and a thermogram generation module 14.

[0074] The 3D-MRI acquisition module 11 is used to acquire three-dimensional structural images of the skull. Preferably, a 1.5T or 3.0T MRI scanner is used, employing the 3DFSET1 high-resolution sequence, with the repetition time (TR) set to 8–12 seconds, the echo time (TE) set to 3–5 ms, the slice orientation field of view (FOV) set to 256–280 mm, the matrix center FOV set to 256–280 mm, the phase FOV set to 0–80%, and the flip angle set to 13–18°. These parameter settings ensure the acquisition of high-resolution structural images, providing a good foundation for subsequent cortical reconstruction.

[0075] The preprocessing module 12 is connected to the 3D-MRI acquisition module 11 and is used to remove the influence of skull, dura mater, etc. on the acquired 3D-MRI. Specifically, the recon-all command in the FreeSurfer software package is used to perform head resampling, bone stripping, intensity normalization, skin stripping, non-brain tissue stripping, white matter / gray matter / cerebrospinal fluid segmentation, etc., to remove non-brain tissue structures that affect cortical reconstruction.

[0076] The cortical reconstruction module 13 is connected to the preprocessing module 12 and is used to automatically measure cortical thickness using the Laplacian manifold reconstruction algorithm. The Laplacian manifold reconstruction algorithm is one of the core innovations of this invention. This algorithm achieves accurate measurement of cortical thickness by solving the distribution of the Laplace equation on the cortex. The core steps of the algorithm are as follows:

[0077] 1. First, the preprocessed 3D-MRI data is converted into a voxel grid representation, where each voxel represents a point in space and its value represents the signal intensity at that point.

[0078] 2. Next, by applying an edge detection algorithm to the voxel grid, the interface between white matter and gray matter (inner boundary) and the interface between gray matter and cerebrospinal fluid (outer boundary) are identified.

[0079] 3. Then, construct a manifold mesh model, representing the inner and outer boundaries as triangular mesh surfaces.

[0080] 4. The mathematical expression for calculating a Laplacian manifold is:

[0081] ,

[0082] in, Represents the Laplace operator. Indicates at point The potential field function at that location. In cortical thickness measurement, It can be understood as a distance function from the inner boundary to the outer boundary.

[0083] 5. Set boundary conditions on the inner boundary. Set boundary conditions on the outer boundary. .

[0084] 6. Solve the Laplace equation above to obtain the values ​​for the entire cortical region. Value distribution.

[0085] 7. Generate streamlines from the inner boundary to the outer boundary, which satisfy the following equations:

[0086] ,

[0087] in, Represents a position vector in space. Indicates parameters, express The gradient.

[0088] 8. Integrate along the streamlines to calculate the skin thickness:

[0089] ,

[0090] in, Point The cortical thickness at that point, the integral along the path through the point The streamlined process proceeds.

[0091] The algorithm described above can accurately measure the thickness of the dermis, achieving a high precision of 0.1 mm.

[0092] The heatmap generation module 14 is connected to the cortical reconstruction module 13 and is used to generate a cortical thickness heatmap with an accuracy of 0.1 mm. The heatmap uses color coding to represent cortical thickness, typically using a red-yellow-green-blue scale, where red represents the thinnest region of the cortex and blue represents the thickest region. Preferably, the color mapping function of the heatmap uses the following method:

[0093] ,

[0094] in, Point The thickness of the cortex at that location is defined by Normalized, which is a normalization function that maps the thickness value to the range (0,1), and ColorMap, which is a color mapping function that maps the values ​​in the range [0,1] to the corresponding colors.

[0095] like Figure 3 As shown, the fNIRS blood flow data acquisition unit 2 includes: a signal acquisition module 21, a coordinate resampling module 22, and a signal preprocessing module 23.

[0096] The signal acquisition module 21 is used to measure skin blood flow signals in 1024 channels in the child's brain via an fNIRS helmet during resting-state fNIRS blood flow signal acquisition. Preferably, a high-density fNIRS helmet covering the entire brain is used, with a sampling rate set to 10Hz and light source wavelengths set to 760nm and 850nm, respectively. These two wavelengths correspond to the absorption peaks of deoxyhemoglobin and oxyhemoglobin, effectively distinguishing between the two states of hemoglobin. Each channel includes a transmitter and a receiver, with a distance of 3cm between them to ensure that the measurement depth reaches the cortex. The acquisition time is set to 5 minutes to obtain sufficient resting-state data.

[0097] The coordinate resampling module 22 is connected to the signal acquisition module 21 and is used to resample the signal in the brain structural coordinate system and spatially match the blood flow data using cortical reconstruction data to calculate the blood flow signal on the cortex. Specifically, firstly, the position of the fNIRS sensor is recorded using a 3D positioning system, and then spatially registered with the child's 3D-MRI structural image to establish the correspondence between the sensor position and the cerebral cortex structure. Next, the propagation path of light in the head tissue is calculated using a light propagation model (such as an improved diffusion approximation model) to determine the cortical region measured by each channel. Finally, the blood flow signal is mapped to the corresponding cortical region to generate a blood flow signal distribution covering the entire cortex.

[0098] The signal preprocessing module 23 is connected to the coordinate resampling module 22 and is used to preprocess the blood flow signal. It applies a sliding window filtering algorithm to smooth periodic noise, calculates the blood oxygen dependence level, and marks areas with a blood oxygen dependence change rate less than -10% as damaged areas. The implementation of the sliding window filtering algorithm is as follows:

[0099] 1. Set the sliding window size w (usually 20-30 data points, corresponding to 2-3 seconds of data).

[0100] 2. For each point i in the time series, calculate the mean within the window:

[0101] ,

[0102] in, This represents the blood flow signal value at time point j. This represents the average value within the window at time point i.

[0103] 3. Calculate the standard deviation within the window:

[0104] ,

[0105] in, This represents the standard deviation within the window at time point i.

[0106] 4. Filtered signal The calculation is as follows:

[0107] ,

[0108] This method can effectively remove sudden noise while preserving the true changes in the signal.

[0109] The calculation of blood oxygen dependence level is based on the modified Beer-Lambert law:

[0110] ,

[0111] ,

[0112] in, This indicates changes in the concentration of oxyhemoglobin. This indicates changes in the concentration of deoxyhemoglobin. and These represent the absorbance changes at wavelengths of 760 nm and 850 nm, respectively. , , and These represent the extinction coefficients of oxyhemoglobin and deoxyhemoglobin at wavelengths of 760 nm and 850 nm, respectively. Indicates the distance between the light source and the detector This represents the differential path factor.

[0113] The blood oxygen-dependent rate of change is calculated as follows:

[0114] ,

[0115] in, Indicates the rate of change in blood oxygen dependence. This indicates changes in the concentration of oxyhemoglobin. This indicates the concentration of oxyhemoglobin at baseline.

[0116] Areas with a blood oxygen-dependent change rate less than -10% were marked as damaged areas. The choice of -10% as the threshold was based on extensive clinical research data, which effectively distinguishes between normal fluctuations and pathological changes. Under normal physiological conditions, fluctuations in blood oxygen-dependent levels are typically within ±5%, while a sustained decrease of more than 10% usually indicates insufficient tissue perfusion or impaired oxygen utilization.

[0117] like Figure 4 As shown, the multi-center data normalization unit 3 includes: a data acquisition module 31, a region segmentation module 32, an anomaly judgment module 33, a consistency analysis module 34, and a damage region generation module 35.

[0118] The data acquisition module 31 is used to acquire data on cortical thinning and decreased blood oxygenation. Preferably, cortical thickness and blood oxygenation level data of at least 100 children with hypoxic-ischemic brain injury are acquired from different centers. These data need to include cases of different ages (newborns, infants, and toddlers) and different degrees of severity to ensure the generalization ability of the model.

[0119] The region segmentation module 32 is connected to the data acquisition module 31 and is used to segment the thinned cortical regions according to cortical thickness atlases. Specifically, the cerebral cortex is divided into multiple regions using anatomical standards, such as the frontal lobe, parietal lobe, temporal lobe, and occipital lobe, or more detailed segmentation methods such as Brodmann partitioning or Desikan-Killiany partitioning are used. For each region, the mean cortical thickness and standard deviation of normal children are calculated to establish region-specific normal ranges.

[0120] The anomaly detection module 33 is connected to the region segmentation module 32. It is used to define the range of brain structural abnormalities according to a standard threshold Z-score > 3 and to mark these points. The formula for calculating the Z-score is as follows:

[0121] ,

[0122] in, Represents the Z-score value. This indicates the cortical thickness value to be evaluated. This indicates the average cortical thickness of normal children in this area. It represents the standard deviation.

[0123] Choosing a Z-score greater than 3 as the threshold for anomaly detection is based on statistical principles. In a normal distribution, the probability of a Z-score greater than 3 is less than 0.3%, meaning that such changes in thickness are extremely rare in the normal population and can therefore be considered abnormal. At the same time, this threshold balances sensitivity and specificity, avoiding excessive false positives or false negatives.

[0124] The consistency analysis module 34 is connected to the anomaly detection module 33 and is used to analyze all data to achieve consistency of multi-center data and establish a unified cortical thickness map analysis system. Specifically, the following steps are used to achieve consistency of multi-center data:

[0125] 1. Calibrate MRI equipment from different centers, establish conversion relationships between equipment, and eliminate systematic errors caused by equipment differences.

[0126] 2. A common anatomical template and segmentation algorithm are used to ensure that the cortical reconstruction results from different centers are comparable.

[0127] 3. Standardize the data from each center to eliminate center-specific bias:

[0128] ,

[0129] in This represents the standardized value. Represents the original value. and Let these represent the mean and standard deviation of the center, respectively. and These represent the mean and standard deviation of the standard reference population, respectively.

[0130] 4. Multicenter data are integrated through meta-analysis to generate comprehensive normal ranges and anomaly judgment criteria.

[0131] The damage area generation module 35 is connected to the consistency analysis module 34 and is used to generate cortical damage areas. Based on standardized data and a Z-score > 3 criterion, abnormal cortical areas are marked, and a three-dimensional model of the damage area is generated. Preferably, the damage area is represented using a triangular mesh model, where each vertex of a triangle corresponds to a point on the cortical surface, and the color code of the vertex corresponds to the Z-score value of that point to visually display the severity of the abnormality.

[0132] like Figure 5 As shown, the sensitivity value selection unit 4 includes: a topological space construction module 41, a spatial registration module 42, a mapping quality evaluation module 43, an intersection region extraction module 44, a distance metric module 45, an intersection region optimization module 46, a sensitivity value candidate set generation module 47, a prior knowledge modeling module 48, and a posterior probability calculation module 49.

[0133] The topology space construction module 41 is used to convert cortical thickness data into a structural topology space, convert blood oxygenation level data into a functional topology space, and establish a homeomorphic mapping relationship between the two spaces. Specifically, the construction methods for the structural and functional topology spaces are as follows:

[0134] For the structural topological space, the skin surface is first represented as a triangular mesh. ,in Represents a set of vertices. Denotes the set of edges. Represents a set of faces. Each vertex This corresponds to a point on the surface of the cortex and assigns it a thickness value. Define the structural topology space. ,in Map each vertex to its thickness value.

[0135] For the functional topology space, triangular meshes are also constructed. Represents the cortical surface, each vertex This corresponds to a point on the cortical surface and is assigned a blood oxygen-dependent level value. Define the functional topology space. ,in Map each vertex to its blood oxygen dependence level value.

[0136] Spatial registration module 42 is connected to topological space construction module 41 to establish a unified brain standard spatial coordinate system. An elastic registration method is applied to align the two topological spaces, constructing a point-to-point mapping relationship with an accuracy controlled within 0.5 mm. The elastic registration method is based on a deformation field model and is achieved by solving the following energy minimization problem:

[0137] ,

[0138] in, Represents the topological space of the structure To functional topology space The mapping function, Represents data items and measures the similarity after mapping. This represents the smoothing term, ensuring the smoothness of the deformation field. This represents the weighting coefficient of the smoothing term.

[0139] Data items are defined as follows:

[0140] ,

[0141] in, and These represent the feature vectors (such as curvature, direction, etc.) of the structural space and functional space, respectively.

[0142] The smoothing term is defined as:

[0143] ,

[0144] in, Represents vertices The set of neighbors.

[0145] The above energy minimization problem is solved by iterative optimization algorithms (such as gradient descent) to obtain the optimal mapping function. Establish a point-to-point correspondence between two topological spaces.

[0146] The mapping quality evaluation module 43 is connected to the spatial registration module 42 and is used to design a mapping quality evaluation index system to evaluate point correspondence accuracy, regional preservation, and topological consistency, and to iteratively optimize the mapping parameters. Specifically, the following three evaluation indices are designed:

[0147] 1. Point-to-point accuracy index:

[0148] ,

[0149] in, Indicates and The actual location of the corresponding anatomical landmarks. Indicates the mapped position. This represents the total number of vertices. The smaller this value, the higher the accuracy of the point correspondence.

[0150] 2. Regional Preservation Indicators:

[0151] ,

[0152] in, Represents a triangle The area of ​​the region. The smaller this indicator is, the smaller the deformation of the region.

[0153] 3. Topology consistency metrics:

[0154] ,

[0155] in, This represents the set of edges that maintain the same topological connections before and after the mapping. The closer this metric is to 1, the better the topology preservation.

[0156] Based on the above indicators, an iterative optimization strategy is adopted to adjust the registration parameters (such as the λ value) until the preset quality threshold or the maximum number of iterations is reached.

[0157] The intersection region extraction module 44 is used to calculate the overlap between the cortical thinning region and the blood oxygenation abnormality region in a unified coordinate space, and to identify significantly related regions by applying spatial correlation analysis to construct a preliminary intersection region representation. Specifically, the cortical thinning region is first defined. and areas with abnormal blood oxygenation :

[0158] ,

[0159] ,

[0160] in, This represents the cortical thickness threshold (usually set to 2.5 mm). This represents the threshold for blood oxygen dependence (usually set to -10% of the normal value).

[0161] Then, calculate the intersection of these two regions:

[0162] ,

[0163] To assess the significance of the intersection region, the spatial correlation coefficient is calculated:

[0164] ,

[0165] in, and These represent the average values ​​of cortical thickness and oxygen-dependent level, respectively. Point The blood oxygen-dependent level at the location.

[0166] Significantly correlated regions are identified by threshold screening (usually r < -0.5, indicating a strong negative correlation).

[0167] The distance metric module 45 is connected to the intersection region extraction module 44 to calculate the Hausdorff distance between regions, quantify region similarity, and design a partitioning calculation strategy to handle irregularly shaped regions. The Hausdorff distance is calculated as follows:

[0168] ,

[0169] in,

[0170] ,

[0171] ,

[0172] For irregularly shaped regions, a partitioned calculation strategy is adopted: first, the region is divided into multiple sub-regions; then, the Hausdorff distance between each pair of sub-regions is calculated separately; finally, a weighted average is taken as the overall distance. The weights are proportional to the areas of the sub-regions.

[0173] The intersection region optimization module 46 is connected to the distance metric module 45. It is used to optimize the intersection boundary using a region growing algorithm, introduce morphological processing to eliminate noise and small holes, and construct a hierarchical representation of the intersection region. The implementation of the region growing algorithm is as follows:

[0174] 1. From the initial intersection area Seed points are selected from the data (usually points with the lowest thickness value and the lowest blood oxygen level).

[0175] 2. Initialize the growth region .

[0176] 3. Regarding Each point in Examine its neighboring points For each neighboring point Add it if the following conditions are met. :

[0177] Not here middle;

[0178] ,

[0179] ;

[0180] in, and These are tolerance parameters for thickness and blood oxygen level, used to control the extent of regional growth.

[0181] 4. Repeat step 3 until no new points are added to G.

[0182] Then, morphological processing (such as opening and closing operations) is applied to eliminate noise and small holes:

[0183] ,

[0184] in, Indicates corrosion operation. This indicates an expansion operation. This represents a structural element (usually a spherical structural element with a radius of 2-3 vertices).

[0185] Finally, a hierarchical representation of the intersection region is constructed, dividing the region into core area, transition area, and edge area:

[0186] Core area: meets and The set of points;

[0187] Transition zone: meets the requirements and The point set; edge region: satisfying and The set of points;

[0188] The sensitivity value candidate set generation module 47 is used to determine the initial range of sensitivity values ​​[0.7, 0.95] based on historical data analysis, and to design a uniform sampling strategy to generate a sensitivity value candidate set. The sensitivity value represents the reliability threshold of the intersection region, used to screen the final responsibility area. Based on extensive clinical data analysis, the sensitivity values ​​typically fall within the range of [0.7, 0.95]. A uniform sampling strategy is used to generate the candidate set with a step size of 0.05: {0.7, 0.75, 0.8, 0.85, 0.9, 0.95}.

[0189] The prior knowledge modeling module 48 is connected to the sensitivity value candidate set generation module 47, and is used to integrate clinical expert knowledge to construct a prior distribution, adjusting the prior based on factors such as patient age and injury type. Specifically, the prior distribution adopts a Beta distribution:

[0190] ,

[0191] in, Indicates the sensitive value. and It is a distribution parameter. This is a Beta function. Based on clinical experience, the initial settings... and The expected value of the prior distribution is This reflects the clinical experience that the sensitivity value is usually around 0.9.

[0192] Adjustment for patients of different ages and value:

[0193] Newborn: (Expected value 0.89);

[0194] baby: (Expected value 0.9);

[0195] Toddlers: (Expected value 0.91);

[0196] This adjustment reflects the impact of differences in brain development among children of different ages on the selection of sensitivity values.

[0197] The posterior probability calculation module 49 is connected to the prior knowledge modeling module 48. It updates the sensitivity value probability distribution based on observed data, applies an iterative optimization algorithm to maximize the posterior probability, verifies the consistency of the sensitivity value across different subsets, and ultimately determines a sensitivity value of 0.9 as the optimal threshold. The posterior probability calculation is based on Bayes' theorem:

[0198] ,

[0199] in, This represents the posterior probability of the sensitivity value θ given data D. Represents the likelihood function. This represents the prior probability.

[0200] The likelihood function is constructed based on the consistency between the region of responsibility and the clinical outcome.

[0201] ,

[0202] in, This represents the clinical outcome of the i-th patient. This represents the area of ​​responsibility determined using the sensitivity value θ, where n represents the total number of patients.

[0203] The optimal sensitivity value is determined by maximizing the posterior probability.

[0204] ,

[0205] To verify the stability of the sensitivity value, a k-fold cross-validation method (usually k=5 or k=10) was used to repeat the above calculation process on different data subsets to check the consistency of the sensitivity value. Practice showed that the sensitivity value of 0.9 exhibited good consistency across different data subsets, and therefore it was determined as the final sensitivity value.

[0206] like Figure 6 As shown, the responsibility area generation unit 5 includes: a boundary precise positioning module 51, a hierarchical boundary representation module 52, a boundary characteristic analysis module 53, a dynamic characteristic analysis module 54, a multi-dimensional feature space construction module 55, a responsibility area hierarchical classification module 56, and a responsibility area visualization module 57.

[0207] The boundary precision localization module 51 is used to enhance the boundary of the responsibility area by applying an edge detection algorithm, introducing multi-directional gradient analysis to improve boundary continuity, and applying sub-pixel boundary localization technology to improve accuracy, generating a high-precision boundary representation with an accuracy of 0.2mm. The edge detection algorithm adopts an improved Canny algorithm, combined with multi-scale analysis to improve detection accuracy.

[0208] 1. Apply Gaussian smoothing filter to the candidate regions of responsibility:

[0209] ,

[0210] in,, Represents the original image. The standard deviation is expressed as The Gaussian kernel is used, and * indicates the convolution operation.

[0211] 2. Calculate the image gradient:

[0212] ,

[0213] 3. Calculate the gradient magnitude and direction:

[0214] ,

[0215] ,

[0216] 4. Non-maximum suppression: If the gradient magnitude at the current point is not a local maximum along the gradient direction, then set it to 0.

[0217] 5. Dual threshold processing: Set a high threshold and low threshold (generally ), the gradient magnitude is greater than Points with gradient magnitudes between [a certain value] are marked as strong edge points. and Points between these points are marked as weak edge points.

[0218] 6. Edge Connection: Starting from a strong edge point, check the adjacent weak edge points along the 8-connected region. If a weak edge point is connected to a strong edge point, mark it as an edge point as well.

[0219] Multidirectional gradient analysis improves boundary continuity by calculating gradients in multiple directions and then fusing the results.

[0220] ,

[0221] in, To represent the direction angle, typically eight evenly distributed directions are chosen: .

[0222] Subpixel boundary positioning technology achieves a high accuracy of 0.2mm through interpolation methods:

[0223] ,

[0224] ,

[0225] The hierarchical boundary representation module 52 is connected to the boundary precision positioning module 51, and is used to construct a multi-level boundary representation of the responsibility area, defining the core area, transition area, and edge area, and assigning reliability weights to different levels of areas. Specifically, based on the degree of abnormality in cortical thickness and blood oxygenation level, the responsibility area is divided into three levels:

[0226] 1. Core region: The area with the thinnest cortical thickness and the lowest blood oxygenation level, typically characterized by a cortical thickness Z-score > 5 and a blood oxygenation-dependent variability rate < -15%. These regions are assigned the highest reliability weight, usually 0.9-1.0.

[0227] 2. Transition zone: A region where the cortical thickness decreases moderately and the blood oxygenation level decreases moderately, typically characterized by a cortical thickness Z-score between 3 and 5 and a blood oxygenation-dependent change rate between -10% and -15%. These regions are assigned a moderate reliability weight, typically 0.7–0.9.

[0228] 3. Marginal Zone: Areas where the cortical thickness is slightly thinner or the blood oxygen level is slightly reduced, typically characterized by a cortical thickness Z-score between 2 and 3 or a blood oxygen-dependent change rate between -5% and -10%. These areas are assigned a lower reliability weight, usually 0.5-0.7.

[0229] The boundary characteristic analysis module 53 is connected to the layered boundary representation module 52. It is used to analyze the geometric characteristics and curvature distribution of the boundary, identify feature points and key segments of the boundary, and evaluate the relationship between the boundary and the anatomical structure. Geometric characteristic analysis includes calculating basic features such as the boundary's length, area, and perimeter, as well as complex features such as shape index and compactness.

[0230] ,

[0231] Where A represents area and P represents perimeter. The closer the shape index is to 1, the closer the shape is to a circle.

[0232] Curvature distribution analysis identifies high-curvature regions (such as sharp corners) and low-curvature regions (such as smooth segments) by calculating the curvature at each point on the boundary:

[0233] ,

[0234] Where κ(s) represents the curvature at the point with parameter s, (x(s), y(s)) represents the coordinates of that point, and x'(s), y'(s), x''(s), and y''(s) represent the first and second derivatives of x(s) and y(s) with respect to s, respectively.

[0235] Curvature analysis identifies characteristic points (such as curvature extrema) and key segments (such as high curvature segments) on the boundary, which are of great significance for the location and evaluation of the area of ​​responsibility.

[0236] The dynamic characteristic analysis module 54 is used to analyze the temporal dynamic characteristics of the blood oxygenation signal within the responsibility area, extract the frequency, amplitude, and phase features of the signal, and identify abnormal dynamic patterns. Specifically, it performs time-domain and frequency-domain analysis on the blood oxygenation signal within the responsibility area:

[0237] 1. Time domain analysis: Calculate the statistical characteristics of a signal, such as mean, standard deviation, skewness, kurtosis, etc.

[0238] 2. Frequency Domain Analysis: The signal is converted to the frequency domain using Fast Fourier Transform (FFT) to analyze its spectral characteristics. For a blood oxygenation signal with a sampling rate of 10Hz, the focus is typically on the spectrum in the range of 0.01-0.5Hz, which corresponds to fluctuations related to heart rate, respiration, and neural activity.

[0239] 3. Time-Frequency Analysis: This involves analyzing the characteristics of a signal at different time and frequency scales using methods such as wavelet transform. The formula for wavelet transform is as follows:

[0240] ,

[0241] in, Let f(t) represent the wavelet coefficients of the function f(t) at scale a and position b, ψ represent the wavelet function, and ψ* represent its complex conjugate.

[0242] These analyses identify abnormal dynamic patterns, such as abnormally reduced amplitude, abnormal frequency shifts, or abnormal phase delays, which are correlated with the severity and type of brain injury.

[0243] The multidimensional feature space construction module 55 is connected to the dynamic characteristic analysis module 54. It integrates structural, functional, and dynamic characteristics to construct a multidimensional feature space, design feature normalization strategies, and build a responsibility region feature fingerprint. Specifically, it integrates the various features obtained from the aforementioned analysis into a single multidimensional feature vector.

[0244] ,

[0245] in, Indicates the first The features include structural features (such as cortical thickness, Z-score, etc.), functional features (such as blood oxygenation level, rate of change, etc.), and dynamic features (such as frequency, amplitude, phase, etc.). To ensure comparability of features at different scales and units, a feature normalization strategy is adopted:

[0246] ,

[0247] in, Represents the normalized features. and Representing features respectively The mean and standard deviation.

[0248] Based on the normalized feature vector, a feature fingerprint of the responsibility area is constructed. This is a visual representation, usually in the form of a radar chart or heatmap, which intuitively displays the multidimensional features of the responsibility area.

[0249] The responsibility classification module 56 is connected to the multi-dimensional feature space construction module 55, and is used to design a responsibility severity assessment system, classify responsibilities based on feature fingerprints, and construct a classification standard library. Specifically, the following classification system is designed:

[0250] 1. Mild injury: Cortical thickness Z-score between 3 and 4, blood oxygen dependence change rate between -10% and -15%, and area of ​​responsibility less than 5 cm².

[0251] 2. Moderate injury: Cortical thickness Z-score between 4 and 5, blood oxygen dependence change rate between -15% and -20%, and area of ​​responsibility between 5 and 10 cm².

[0252] 3. Severe injury: Cortical thickness Z-score greater than 5, blood oxygen-dependent change rate less than -20%, and area of ​​responsibility greater than 10 cm².

[0253] Based on the characteristic fingerprints and the aforementioned grading criteria, each area of ​​responsibility is classified. Simultaneously, a grading standard library is constructed, containing the characteristic fingerprints of numerous typical cases and their corresponding grading results, serving as a reference for grading new cases.

[0254] The responsibility area visualization module 57 is connected to the responsibility area hierarchical classification module 56, and is used to design a multimodal fusion visualization scheme. It employs color coding to represent different characteristics and constructs an interactive visualization interface. Specifically, the following visualization strategies are adopted:

[0255] 1. 3D Surface Rendering: The area of ​​responsibility is rendered as a three-dimensional surface on the cortical model, with different colors representing different degrees of severity or different layered areas.

[0256] 2. Fault visualization: Provides coronal, sagittal and transverse fault visualization to facilitate observation of the depth distribution of the area of ​​responsibility.

[0257] 3. Multimodal overlay: Overlaying structural images (such as T1-weighted images) and functional images (such as blood oxygen level images) to intuitively reflect the structure-function relationship.

[0258] 4. Interactive functions: Provides basic interactive functions such as rotation, scaling, and translation, as well as advanced functions such as feature filtering and hierarchical display to meet different clinical needs.

[0259] Through the visualization solutions described above, clinicians can intuitively understand the location, extent, severity, and characteristic distribution of the area of ​​responsibility, providing support for diagnostic and treatment decisions.

[0260] like Figure 10 As shown, the method for analyzing cortical thickness maps in hypoxic-ischemic brain injury according to the present invention includes the following steps:

[0261] Step 1: Acquire 3D-MRI and perform cortical reconstruction. Automatically measure cortical thickness using the Laplacian manifold reconstruction algorithm to generate a cortical thickness heatmap with an accuracy of 0.1 mm.

[0262] Step 2: Obtain fNIRS blood flow signals from the cerebral cortex. Measure 1024 channels of skin blood flow signals in the child's brain using an fNIRS helmet. Resample the signals using the brain structural coordinate system and spatially match the blood flow data using cortical reconstruction data. Calculate the blood flow signals on the cortex, preprocess the blood flow signals, apply a sliding window filtering algorithm to smooth periodic noise, calculate the blood oxygen dependence level, and mark areas with a blood oxygen dependence change rate less than -10% as damaged areas.

[0263] Step 3: Using cortical thickness data, construct a multicenter dataset and analyze the multicenter dataset. Support the normalization of the multicenter dataset and identify the parts with Z-score > 3 as outliers.

[0264] Step 4: Based on the topological space intersection mapping technique, the cortical thinning region and the abnormal blood oxygenation signal region are superimposed. The cortical thickness data is converted into structural topological space, and the blood oxygenation level data is converted into functional topological space. A homeomorphic mapping relationship is established between the two spaces. The overlap between the cortical thinning region and the abnormal blood oxygenation region is calculated in a unified coordinate space. Spatial correlation analysis is applied to identify significantly related regions. The Hausdorff distance between regions is calculated to quantify the similarity of regions. The intersection region of the two is selected as the damage responsibility area. The sensitivity value is determined to be 0.9 through sensitivity value analysis, and the damage responsibility area is finally determined.

[0265] Step 5: Based on the abnormalities in the cortex, apply edge detection algorithms to enhance the boundaries of the responsibility area, construct a multi-level boundary representation of the responsibility area, analyze the geometric characteristics and curvature distribution of the boundaries, analyze the temporal dynamic characteristics of blood oxygenation signals within the responsibility area, integrate structural, functional, and dynamic characteristics, construct a multi-dimensional feature space, and generate a responsibility area map and responsibility area sensitivity values.

[0266] The above steps correspond one-to-one with each functional unit of the system, and the specific implementation methods have been described in detail in the system section.

[0267] The application effect of the present invention will be illustrated below through a specific embodiment.

[0268] Taking a 6-month-old infant with hypoxic-ischemic brain injury as an example, the system and method of this invention were analyzed. First, 3D-MRI data of the infant's head were acquired using a 3.0T MRI scanner. The scanning parameters were set as follows: TR=10 seconds, TE=4 ms, slice direction FOV=260 mm, matrix center FOV=260 mm, phase FOV=70%, and flip angle=15°. After acquisition, preprocessing was performed using the FreeSurfer software package to remove non-brain tissues such as skull and dura mater. Then, the Laplacian manifold reconstruction algorithm was applied to reconstruct the cortex and measure its thickness, generating a cortical thickness thermogram with an accuracy of 0.1 mm.

[0269] Next, high-density fNIRS helmets were used to acquire cerebral blood flow signals from the children, obtaining signal data from 1024 channels at a sampling rate of 10 Hz for 5 minutes. Spatial registration and coordinate resampling were performed on the acquired signals, mapping them to the same coordinate system as 3D-MRI. Then, a sliding window filtering algorithm (window size of 25 data points) was applied to preprocess the signals, removing physiological noise and motion artifacts. Based on the modified Beer-Lambert law, oxygen dependence levels were calculated, and areas with an oxygen dependence change rate less than -10% were marked as potential injury areas.

[0270] Subsequently, the cortical thickness data of the children were compared with a multicenter database containing 120 age-matched children, and abnormal areas were identified using Z-score calculation. The analysis revealed that the cortical thickness in the left frontoparietal lobe region of the children was significantly thinner, with a Z-score of 4.2, which significantly exceeded the abnormality judgment threshold of 3.

[0271] Next, the topological intersection mapping technique was applied to fuse and analyze the thinning cortical region with the abnormal blood oxygenation signal region. First, structural and functional topological spaces were constructed, both represented by triangular meshes with approximately 150,000 vertices. Then, a homeomorphic mapping between the two spaces was established using an elastic registration method, achieving a registration accuracy of 0.3 mm. The mapping quality was evaluated, with a point correspondence accuracy of 0.28 mm, a region preservation index of 0.05, and a topological consistency index of 0.97, all meeting the quality requirements.

[0272] In a unified coordinate space, the overlap between thinned cortical regions (thickness < 2.5 mm) and abnormal blood oxygenation regions (change rate < -10%) was calculated. Significant overlap was found in the left frontoparietal lobe region, with a spatial correlation coefficient r = -0.68, indicating a strong negative correlation. Further quantification of regional similarity using Hausdorff distance yielded a distance value of 2.3 mm, below the threshold of 3 mm, confirming a high degree of similarity between the two regions.

[0273] Region growing algorithms and morphological processing are applied to optimize the boundaries of intersection regions, eliminating noise and small holes. Then, through Bayesian sensitivity estimation, combined with prior knowledge (infant age group, prior distribution parameters α=9, β=1) and observed data, the posterior probability distribution is calculated, and the optimal sensitivity value is determined to be 0.9. The stability of the sensitivity value is verified by 5-fold cross-validation, with a standard deviation of only 0.02 across different data subsets.

[0274] Finally, the improved Canny edge detection algorithm was applied to accurately locate the boundary of the responsibility area, achieving a boundary localization accuracy of 0.2 mm. A multi-level representation of the responsibility area was constructed, identifying the core area (2.8 cm²), transition area (1.5 cm²), and edge area (1.2 cm²). Analysis of the boundary's geometric characteristics yielded a shape index of 0.73, indicating a relatively irregular shape for the responsibility area. Curvature analysis identified three high-curvature feature points located at the boundary between the responsibility area and important functional areas.

[0275] Analysis of the dynamic characteristics of blood oxygenation signals within the area of ​​responsibility revealed a significant decrease in frequency components within the 0.05-0.1 Hz range, with amplitude reduced by 46% compared to the normal area, suggesting significant inhibition of neural activity. Integrating structural, functional, and dynamic features, a characteristic fingerprint of the area of ​​responsibility was constructed, and based on this fingerprint, the area was classified as moderately damaged (cortical thickness Z-score = 4.2, blood oxygenation-dependent change rate = -16%, total area of ​​the area of ​​responsibility = 5.5 cm²).

[0276] Ultimately, a 3D visualization of the area of ​​responsibility is generated, with different colors encoding different layered areas (red for the core area, yellow for the transition area, and green for the edge area), and multi-angle and tomographic displays are provided. Clinicians can use this visualization to accurately locate the area of ​​responsibility for the injury and develop targeted rehabilitation treatment plans accordingly. Six-month follow-up results showed significant improvement in the child's neurodevelopmental indicators, validating the accuracy and clinical value of the system's analysis.

[0277] As can be seen from the above embodiments, the hypoxic-ischemic brain injury cortical thickness atlas analysis system and method of the present invention, by fusing high-precision structural imaging and functional imaging data and applying topological space intersection mapping technology, realizes the accurate localization and assessment of the brain injury responsibility area, providing important support for clinical diagnosis and treatment.

Claims

1. A system for cortical thickness mapping of hypoxic-ischemic brain injury, characterized by, The method comprises the following steps: A cortical thickness data acquisition unit is used to acquire 3D-MRI and perform cortical reconstruction, automatically measure the cortical thickness by a Laplacian manifold reconstruction algorithm, and generate a cortical thickness heat map with a precision of 0.1 mm; An fNIRS blood flow data acquisition unit is used to acquire the fNIRS blood flow signal of the cerebral cortex, perform blood oxygen-dependent level preprocessing, and generate a blood flow signal heat map; A multi-center data normalization unit is connected with the cortical thickness data acquisition unit and the fNIRS blood flow data acquisition unit, used to construct multi-center data by using the cortical thickness data, analyze the multi-center data, support the normalization of the multi-center data, and judge the part with a Z-score greater than 3 as abnormal; A sensitivity value selection unit is connected with the multi-center data normalization unit, used to superimpose the cortical thinning area and the blood oxygen signal abnormal area based on a topological space intersection mapping technology, select the intersection area of the two as the damage responsibility area, determine the sensitivity value 0.9 through sensitivity value analysis, and finally determine the damage responsibility area; A responsibility area generation unit is connected with the sensitivity value selection unit, used to generate a responsibility area map and a responsibility area sensitivity value according to the abnormal condition of the cortex.

2. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The sensitivity value selection unit comprises: A topological space construction module is used to convert the cortical thickness data into a structural topological space, convert the blood oxygen level data into a functional topological space, and establish a homeomorphism mapping relationship between the two spaces; A space registration module is connected with the topological space construction module, used to establish a unified brain standard space coordinate system, align the two topological spaces by applying an elastic registration method, and construct a point-to-point mapping relationship with a precision controlled within 0.5 mm; A mapping quality evaluation module is connected with the space registration module, used to design a mapping quality evaluation index system, evaluate the point correspondence accuracy, area preservation, and topological consistency, and iteratively optimize the mapping parameters.

3. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The sensitivity value selection unit further comprises: An intersection area extraction module is used to calculate the overlap degree of the cortical thinning area and the blood oxygen abnormal area in the unified coordinate space, apply spatial correlation analysis to identify the significantly correlated area, and construct a preliminary intersection area representation; A distance measurement module is connected with the intersection area extraction module, used to calculate the Hausdorff distance between the areas, quantify the area similarity, and design a partition calculation strategy to process irregularly shaped areas; An intersection area optimization module is connected with the distance measurement module, used to apply a region growing algorithm to optimize the intersection boundary, introduce morphological processing to eliminate noise and small holes, and construct a hierarchical representation of the intersection area.

4. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The sensitivity value selection unit further comprises: A sensitivity value candidate set generation module is used to determine the initial range [0.7, 0.95] of the sensitivity value based on historical data analysis, and design a uniform sampling strategy to generate a sensitivity value candidate set; A prior knowledge modeling module is connected with the sensitivity value candidate set generation module, used to integrate clinical expert knowledge to construct a prior distribution, and consider factors such as patient age and injury type to adjust the prior; The posterior probability calculation module is connected with the prior knowledge modeling module, and is configured to update the sensitive value probability distribution based on observation data, maximize the posterior probability by applying an iterative optimization algorithm, verify the consistency of the sensitive value on different subsets, and finally determine the sensitive value 0.9 as the optimal threshold.

5. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The responsibility area generation unit comprises: The boundary accurate positioning module is configured to apply an edge detection algorithm to enhance the responsibility area boundary, introduce multi-directional gradient analysis to improve the boundary continuity, and apply a sub-pixel boundary positioning technology to improve the accuracy, so as to generate a high-precision boundary representation with an accuracy of 0.2 mm; The hierarchical boundary representation module is connected with the boundary accurate positioning module, and is configured to construct a multi-level boundary representation of the responsibility area, define a core area, a transition area and an edge area, and assign reliability weights to different levels of areas; The boundary characteristic analysis module is connected with the hierarchical boundary representation module, and is configured to analyze the geometric characteristics and curvature distribution of the boundary, identify feature points and key segments of the boundary, and evaluate the relationship between the boundary and the anatomical structure.

6. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The cortical thickness data acquisition unit comprises: The 3D-MRI acquisition module is configured to acquire a three-dimensional structure image of the head; The preprocessing module is connected with the 3D-MRI acquisition module, and is configured to remove skull, dura mater and the like from the acquired 3D-MRI; The cortical reconstruction module is connected with the preprocessing module, and is configured to automatically measure the cortical thickness by a Laplacian manifold reconstruction algorithm; The heat map generation module is connected with the cortical reconstruction module, and is configured to generate a cortical thickness heat map with a precision of 0.1 mm.

7. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The fNIRS blood flow data acquisition unit comprises: The signal acquisition module is configured to measure 1024 channels of skin blood flow signals on the brain of the child through an fNIRS helmet when the child is acquiring resting-state fNIRS blood flow signals; The coordinate resampling module is connected with the signal acquisition module, and is configured to resample the signals in a brain structure coordinate system, and perform spatial matching of the blood flow data by using the cortical reconstruction data, and calculate the blood flow signals on the cortex; The signal preprocessing module is connected with the coordinate resampling module, and is configured to preprocess the blood flow signals, apply a sliding window filtering algorithm to smooth periodic noise, calculate blood oxygen-dependent levels, and mark regions with a blood oxygen-dependent rate less than -10% as injury regions.

8. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The multi-center data normalization unit comprises: The data acquisition module is configured to acquire cortical thinning data and blood oxygen reduction data; The region segmentation module is connected with the data acquisition module, and is configured to segment the cortical thinning part according to a cortical thickness atlas; The abnormality judgment module is connected with the region segmentation module, and is configured to divide the range of brain structure abnormalities according to a standard threshold Z-score>3, and mark these points; The consistency analysis module is connected with the abnormality judgment module, and is configured to analyze all data to realize the consistency of multi-center data, and establish a unified cortical thickness atlas analysis system; The injury area generation module is connected with the consistency analysis module, and is configured to generate a cortical injury area.

9. The hypoxic-ischemic brain injury cortical thickness atlas analysis system of claim 1, wherein, The responsibility area generation unit further comprises: a dynamic characteristic analysis module, configured to analyze time dynamic characteristics of the blood oxygen signals in the responsibility area, extract frequency, amplitude, and phase characteristics of the signals, and identify abnormal dynamic patterns; a multi-dimensional feature space construction module, connected with the dynamic characteristic analysis module, configured to integrate structure, function, and dynamic characteristics, construct a multi-dimensional feature space, design a feature normalization strategy, and construct a responsibility area feature fingerprint; a responsibility area classification module, connected with the multi-dimensional feature space construction module, configured to design a responsibility area severity evaluation system, classify responsibility areas based on the feature fingerprint, and construct a classification standard library; a responsibility area visualization module, connected with the responsibility area classification module, configured to design a multi-modal fusion visualization scheme, express different characteristics using color coding, and construct an interactive visualization interface.

10. A method of cortical thickness mapping of hypoxic-ischemic brain injury, characterized by, The method comprises the following steps: 3D-MRI is acquired and cortical reconstruction is performed, cortical thickness is automatically measured by a Laplacian manifold reconstruction algorithm, and a 0.1 mm precision cortical thickness heat map is generated; fNIRS blood flow signals of the brain cortex are acquired, 1024 channels of skin blood flow signals are measured on the brain of the child by an fNIRS helmet, the signals are resampled in a brain structure coordinate system, and the blood flow data are spatially matched by using the cortical reconstruction data, the blood flow signals on the cortex are calculated, the blood flow signals are preprocessed, a sliding window filtering algorithm is applied to smooth periodic noise, blood oxygen dependence levels are calculated, and regions with a blood oxygen dependence change rate less than -10% are marked as injury regions; multi-center data are constructed by using the cortical thickness data, and the multi-center data are analyzed to support normalization of the multi-center data, and regions with a Z-score greater than 3 are judged as abnormal; based on a topological space intersection mapping technology, a cortical thinning region and an abnormal blood oxygen signal region are superimposed, the cortical thickness data are converted into a structural topological space, the blood oxygen level data are converted into a functional topological space, a homeomorphism relationship between the two spaces is established, an overlap degree of the cortical thinning region and the abnormal blood oxygen signal region is calculated in a unified coordinate space, a spatial correlation analysis is applied to identify a significantly correlated region, a Hausdorff distance between the regions is calculated, a region similarity is quantified, an intersection region of the two regions is selected as an injury responsibility area, a sensitivity value of 0.9 is determined through sensitivity value analysis, and finally the injury responsibility area is determined; according to abnormal conditions of the cortex, an edge detection algorithm is applied to enhance a responsibility area boundary, a multi-level boundary representation of the responsibility area is constructed, geometric characteristics and curvature distribution of the boundary are analyzed, time dynamic characteristics of blood oxygen signals in the responsibility area are analyzed, structure, function, and dynamic characteristics are integrated, a multi-dimensional feature space is constructed, and a responsibility area atlas and a responsibility area sensitivity value are generated.

Images